[clang] [llvm] [AMDGPU] Introduce asyncmark/wait intrinsics (PR #173259)

[Matt Arsenault via cfe-commits]

================
@@ -361,6 +361,20 @@ void AMDGPUAsmPrinter::emitInstruction(const MachineInstr *MI) {
       return;
     }
 
+    if (MI->getOpcode() == AMDGPU::S_ASYNCMARK) {
+      if (isVerbose())
+        OutStreamer->emitRawComment(" s_asyncmark");
----------------
arsenm wrote:

Synthetic operations shouldn't get an S_ prefix which makes it look like a real instruction 

https://github.com/llvm/llvm-project/pull/173259